WildCross: A Cross-Modal Large Scale Benchmark for Place Recognition and Metric Depth Estimation in Natural Environments

Joshua Knights; Joseph Reid; Kaushik Roy; David Hall; Mark Cox; Peyman Moghadam

cs.CVMar 2, 2026

WildCross: A Cross-Modal Large Scale Benchmark for Place Recognition and Metric Depth Estimation in Natural Environments

Authors: Joshua Knights, Joseph Reid, Kaushik Roy, David Hall, Mark Cox, Peyman Moghadam

Organizations: CSIRO Robotics, Data61, CSIRO, Australia. · Queensland University of Technology (QUT), Australia.

Abstract

Recent years have seen a significant increase in demand for robotic solutions in unstructured natural environments, alongside growing interest in bridging 2D and 3D scene understanding. However, existing robotics datasets are predominantly captured in structured urban environments, making them inadequate for addressing the challenges posed by complex, unstructured natural settings. To address this gap, we propose WildCross, a cross-modal benchmark for place recognition and metric depth estimation in large-scale natural environments. WildCross comprises over 476K sequential RGB frames with semi-dense depth and surface normal annotations, each aligned with accurate 6DoF poses and synchronized dense lidar submaps. We conduct comprehensive experiments on visual, lidar, and cross-modal place recognition, as well as metric depth estimation, demonstrating the value of WildCross as a challenging benchmark for multi-modal robotic perception tasks. We provide access to the code repository and dataset at https://csiro-robotics.github.io/WildCross.

Figures & tables

	VPR	LPR	CMPR	Depth Est.		ViewPoint	Temporal	Scene
\Block 2-1Name	\Block 1-4Supported Tasks				\Block 1-4Diversity
Nordland [ 1 ]	✓	✗	✗	✗		$\star$	$\star\star\star$	$\star\star\star$
RELLIS-3D [ 2 ]	✓	✓	✓	✗		$\star$	$\star$	$\star$
Wild-Places [ 3 ]	✗	✓	✗	✗		$\star\star\star$	$\star\star\star$	$\star\star\star$
Oxford Forest [ 4 ]	✗	✓	✗	✗		$\star\star\star$	$\star\star$	$\star\star\star$
BotanicGarden [ 5 ]	✓	✓	✓	✗		$\star\star$	$\star$	$\star$

TABLE I : Comparison of existing datasets in natural environments with support for relocalization. Datasets are compared based on the range of supported tasks and their diversity across viewpoint, temporal, and scene variations between revisits.

Panel of Figure 2 — Fig. 2 : WildCross overview (a) RGB Image, (b) Depth Image, (c) Depth Overlay, (d) Surface Normal, (e) Lidar Submap.

\Block 2-2Sequence		\Block 2-1Distance	\Block 1-3Camera / Depth Images			\Block 1-3Submaps
			All	Revisits	%	All	Revisits	%
\Block 4-1Venman	01	2.64km	35.3K	10.6K	30.03	4.7K	1.2K	25.69
	02	2.64km	34.1K	8.3K	24.25	4.6K	980	21.51
	03	4.59km	63.8K	8.7K	13.70	8.5K	906	10.70
	04	2.81km	43.1K	8.0K	18.57	5.7K	937	16.33
\Block 4-1Karawatha	01	5.14km	66.1K	5.6K	8.43	8.8K	553	6.27

TABLE II : Sequence statistics for WildCross . “Revisits” denote intra-sequence loop closures.

Fig. 3 : Impact of visibility estimation. (a) RGB Image, (b) Naïve projection of global 3D points produces noisy depth maps with occluded points. (c) Our visibility pipeline removes these, yielding higher-quality depth.

Fig. 4 : Depth distribution for WildCross ( $\bullet$ ) vs. KITTI Annotated Depth ( $\bullet$ ) [ 35 ] . Violin plots are computed from 1% subsamples of both datasets. Width and height are normalized with respect to image sizes.

\Block 2-1Split	\Block 2-1 V/K-01	\Block 2-1V/K-02	\Block 2-1V/K-03	\Block 2-1V/K-04	\Block 1-2Lidar		\Block 1-2Camera
					Train	Test	Train	Test
01	Test	Train	Train	Train	49.8K	13.5K	374.6K	101.4K
02	Train	Test	Train	Train	48.7K	14.6K	366.3K	109.7K
03	Train	Train	Test	Train	39.7K	23.6K	298.0K	177.9K
04	Train	Train	Train	Test	51.8K	11.5K	389.1K	86.9K

TABLE III : Train/test cross-fold splits for WildCross .

Panel of Table 7 — TABLE IV : Intra-sequence VPR results on WildCross for zero-shot and fine-tuned networks.

Visual Place Recognition

Visual Place Recognition matches camera observations to mapped places despite changes in viewpoint, lighting, weather, or time.

See all 84 papers

Cross-Modal Benchmarking for Robotic Perception in Natural Environments
Jun 10, 2026David Hall, Joshua Knights, Mark Cox +1Metric Depth Estimation Visual Place Recognition
CSIRO Robotics, CSIRO, Australia. · University of Sydney (USyd), Australia. · Queensland University of Technology (QUT), Australia.
WildDepth: A Multimodal Dataset for 3D Wildlife Perception and Depth Estimation
Mar 17, 2026Muhammad Aamir, Naoya Muramatsu, Sangyun Shin +6Depth Estimation 3D Reconstruction
University of Oxford · University of Cape Town · University College London
M3GA-Wild: A Large-Scale Dataset and Benchmark for Multi-Modal Multi-session Ground-to-Aerial Place Recognition in Forests
Sep 19, 2026Ethan Griffiths, Maryam Haghighat, Simon Denman +2Visual Place Recognition LiDAR Place Recognition
Queensland University of Technology (QUT) · CSIRO Robotics

Jun 10, 2026J/K move · Enter open · S save

Cross-Modal Benchmarking for Robotic Perception in Natural Environments

David Hall, Joshua Knights, Mark Cox, Peyman Moghadam

CSIRO Robotics, CSIRO, Australia. · University of Sydney (USyd), Australia. · Queensland University of Technology (QUT), Australia.

Metric Depth Estimation Visual Place Recognition Robotics Robotic Perception

Open paper AskPDF arXiv

Natural environments present a complex challenge to robotics perception systems. Current models, particularly vision foundation models, are largely trained on structured, urban environments leading to weaknesses in their perception for field robotics tasks. We showcase the limitations of current models using our recently released WildCross benchmark, a new cross-modal benchmark for place recognition and metric depth estimation in large-scale natural environments. WildCross comprises over 476K sequential RGB frames with semi-dense depth and surface normal annotations, each aligned with accurate 6DoF pose and synchronized dense lidar submaps. In this work, we provide an expanded analysis of the benchmark results from the recent WildCross benchmark, with particular emphasis on expanded metric depth estimation experiments. Access to the code repository and dataset for this work can be found at https://csiro-robotics.github.io/WildCross.

Abstract

Visual Place Recognition

Explore similar work

Cross-Modal Benchmarking for Robotic Perception in Natural Environments

WildDepth: A Multimodal Dataset for 3D Wildlife Perception and Depth Estimation

M3GA-Wild: A Large-Scale Dataset and Benchmark for Multi-Modal Multi-session Ground-to-Aerial Place Recognition in Forests